MULTIPARTY COLLABORATION IN EDGE COMPUTING SYSTEMS

Abstract

With the recent wave of technological advancements, the deployment of the ``Internet of Things'' (IoT) is becoming ubiquitous, ranging from typical home and personal appliances to sophisticated safety-critical systems such as nuclear plants and medical implants. Complex IoT systems supporting multiple services, particularly edge computing systems, invariably consist of multiple subsystems designed, installed, or managed by a different vendor, organization, or party. In particular, each physical cluster of devices or the services offered on top of these physical clusters may form an independent subsystem focused on a specific service. Each subsystem would have its own independent controller with little knowledge of or access to the data or resources provided by other subsystems. Nevertheless, all of these systems typically operate in the same shared environment and are therefore influenced by the environment and the actions of other controllers. Maintaining a safe and secure operation in such multiparty systems is extremely difficult due to a lack of coordination between subsystems, lack of timely visibility to sensor data streams, and inability to perform, request, or even be aware of actuation by other parties. In addition to smart IoT systems, most of the large systems are composed of subsystems, including edge computing systems which are inherently multiparty, involving multiple vendors at each level, such as devices, edge controllers, virtualization layers, and higher-level services. Edge computing systems introduce the additional problem of limited computing/communication power at the device layer, necessitating the separation of functionality between the device and controller levels and the resulting problems with communication bandwidth and latency. This division of large-scale systems into subsystems that multiple vendors or parties manage forms the main motivation for this dissertation. In this dissertation, we focus on two kinds of subsystems: (1) Smart IoT-based Building Management and (2) Non-intrusive video-based monitoring in Intelligent Transport Systems (ITS). We then address the key issues related to these two subsystems; in the context of ITS, we have developed resource-efficient video-based monitoring of road traffic, detecting and resolving the anomalies and further characterizing driving behaviors. We develop a flexible framework for analyzing the spatiotemporal relationships between events and activities in video-based surveillance. Furthermore, in the context of large-scale smart IoT systems, we have addressed the issues of detecting and resolving policy conflicts and access control issues. The subsystems described above will likely be designed/deployed by different vendors and managed by different administrators, also known as ``parties" ". Therefore, their operational rules (ORs) are developed independently, and these parties must work together to ensure that their ORs do not conflict. We codify the smooth functioning of the entire system through a set of ``safety properties" that must be enforced collaboratively. This dissertation examines two issues that arise in such a multi-party environment: (1) Conflicts between complex policies in multiple subsystems environments that need to be resolved for safe and secure operation and (2) Cross-party access to the state of sensors/actuators and the ability to request remote actuation. To resolve the conflicts arising in multi-party IoT systems, we have developed a model that performs Conflict Detection and Resolution and can determine both Static(compile-time) and Dynamic(run-time) Conflicts. We consider temporal logic operations integral to operational rules and safety properties. Our model, based on intelligent combinatorial optimization, proactively catches potential conflicts and mitigates them by temporarily perturbing the various thresholds or durations in the rules. We also address the problem of access control issues by developing an access control architecture in which we distinguish between the static authorization problem, which selects parties responsible for enforcing safety properties, and the dynamic (run-time) control of accesses. This creates a novel enforcer selection problem, for which we develop efficient algorithms and quantify their performance using a comprehensive smart home emulation. In the context of emerging edge computing applications that use high-definition cameras as edge devices to capture video streams of road traffic that need to be analyzed in real-time for situational understanding and answering queries, we develop an Intelligent Transportation System (ITS) monitoring solution that addresses the critical issue of processing the large amount of data generated in real-time by energy-constrained edge devices. We developed a model for ITS that has two components, A lightweight, energy-efficient video-stream analytics algorithm called YLLO (You Look Less than Once) runs on individual edge devices (EDs). It substantially reduces the video frames sent to the EC, and An algorithm called BATS exploits the overlaps between the views of multiple cameras and dynamically adapts to the available transmission bandwidth to the EC. Within the scope of an ITS (Intelligent Transportation System), we have developed an effective model for (1) Detecting and resolving anomalous events and (2) Assessing driving behavior from on-the-road cameras is also presented in this dissertation. We have developed C-FAR, a framework for reasoning about anomalies based on video monitoring by the roadside camera infrastructure. Unlike traditional approaches that utilize deep learning to recognize individual activities, C-FAR does so only for primitive movements and activities and then builds a comprehensive event logic framework. It also provides an optimal resolution of the detected/predicted anomalies by identifying the minimal changes in the controllable parameters of the system. Furthermore, in this dissertation, we show a proof of concept for characterizing vehicular behavior using only the ITS system's roadside cameras. The primary advantage of this method is that it can be transparently and inexpensively implemented in roadside infrastructure and can provide a global view of each vehicle's behavior without the involvement or awareness of individual vehicles or drivers. The characterization can improve the safety and performance of traffic flow, particularly in mixed manual and automated vehicle scenarios, which are expected to become more common soon.